Checking the answers: did Anthropic's Sonnet 5 go back to its original price on September 1?
From the Oct 5, 2026 daily brief
None of the claims considered for this week's column cleared the bar for opening a tracked entry. So this issue does one thing. It writes up the first three of SecondSource's own calls that have reached a result but that the column hasn't yet covered, in the order they closed. All three closed late. The first came due on September 1 and the other two on September 15. SecondSource checked and closed all three only on September 30.
Would Anthropic's Sonnet 5 go back to its original price of US$3 per million input tokens and US$15 per million output tokens on September 1?
① What the call said then: Anthropic launched Sonnet 5 with an introductory API price of US$2 per million input tokens and US$10 per million output tokens, running through August 31, 2026. SecondSource opened this entry on July 31, meaning it began tracking a call still to be verified. The call as written on its card: "After Anthropic's promotional pricing for the Claude Sonnet 5 API expires on 2026-08-31, the official pricing page returns to the pre-promotion standard rate of $3/M input and $15/M output."
② The actual reading: the source named in advance was Anthropic's official pricing page. Note 3 on that page says the Sonnet 5 price of US$2 input and US$10 output, "announced at launch as introductory pricing through August 31, 2026, is now the standard price." The same note says the increase originally scheduled for September 1 "will not occur." SecondSource checked the page on September 30 and again on October 4, and the text was the same both times (Anthropic pricing page).
③ Ruling, and at which layer: didn't come true, and the failure sits in a premise that events overturned. Anthropic had scheduled the increase. Before the deadline, Anthropic turned the promotional price into the standard price. The call rested entirely on scheduled terms running as written, and it left no room for Anthropic changing its mind. US$2 input and US$10 output is now Anthropic's official standard price for the Sonnet 5 API, not a promotion due to expire. The same entry also shows that Anthropic can change a standard price just as it changed this one.
④ Which tracked judgments moved: none so far. This entry feeds two items on SecondSource's tracking list. One is the judgment that frontier labs can hold on to the value their models create. The other is the frontier-pricing-power trend, which asks whether margins hold once selling tokens turns profitable. This reading should feed a reassessment of both, and that reassessment is still outstanding.
⑤ Does the standard change: Yes. The call treated a scheduled price change as settled, even though a company can reprice before a deadline. From now on, a call that "a scheduled price change will go through" no longer counts as close to a sure thing. "A price change announced before the deadline" is also written in advance into the conditions under which such a call counts as missed.
Would Microsoft's annual report make clear whether OpenAI's credits were counted as Azure revenue?
① What the call said then: the column opened this entry on August 2. The credits in question are cloud credits. Microsoft invested in OpenAI partly in the form of cloud credits, and OpenAI then used those credits to buy Azure services. Bill Gurley of the venture firm Benchmark said Microsoft counted the credits as Azure revenue. Microsoft CEO Satya Nadella said Microsoft did not. The entry put it as "one of them is wrong," and named the accounting notes in Microsoft's FY2026 annual report as the place that could settle it. The entry recorded no confidence level at the time.
② The actual reading: the source named in advance was Microsoft's FY2026 10-K, filed on July 29, 2026 (EDGAR accession number 0001193125-26-323660). The filing was read in the original on September 30. Its notes say "we recorded revenue from commercial arrangements with OpenAI, inclusive of revenue-sharing payments," and put the amount at US$24.1B. The notes give only that total. They don't say how Microsoft booked the credits.
③ Ruling, and at which layer: can't judge. The question here isn't whether the call was right or wrong. The problem lies in the choice of source: the source named in advance can't answer the question. The column opened the entry without first checking whether Microsoft's past annual reports had ever disclosed a figure of this kind.
④ Which tracked judgments moved: none. Gurley's claim and Nadella's claim both stay on the tracking list unchanged. Nothing so far has confirmed or overturned either one. In practice, what Microsoft made public in its FY2026 annual report can't settle the dispute over whether Microsoft counted the credits as Azure revenue.
⑤ Does the standard change: Yes. This entry named a source without checking that the source had ever disclosed this kind of figure. From now on, a source named for an entry must have disclosed the same kind of figure before. If no precedent turns up, the entry names a backup source as well.
Was Anthropic afraid to raise prices, as Gurley said? Did the price go back to standard on September 1, as the terms said?
① What the call said then: this entry reads the same evidence as the first one, but what it corrects is the ruling in the August 16 column. That column ruled that Gurley's claim doesn't hold. Gurley's claim was that nobody dares raise prices, and that Anthropic won't move its prices for fear of losing share. The hardest evidence against Gurley's claim was the official pricing page, which stated that Sonnet 5's introductory price would expire on August 31. So the column opened an entry on whether the price would go back to standard on September 1, as the terms said. That return would have been a 50% increase. The entry recorded no confidence level at the time.
② The actual reading: the same official pricing page and the same note 3 as in the first entry. US$2 input and US$10 output has become the standard price, and the originally scheduled increase will not occur.
③ Ruling, and at which layer: didn't come true, and the failure sits in the reasoning. The August 16 column took "the terms carry an expiry date" as evidence that "the company dares to raise prices." But a company can change its terms before they expire. This was the strongest plank of the August 16 rebuttal of Gurley, and it is now void. The ruling in that column has gone from "doesn't hold" back to "unresolved," and both the August 16 column and that day's daily brief now carry a note saying so.
④ Which tracked judgments moved: on October 3, the column's ruling changed from "doesn't hold" to "unresolved." Three items on the tracking list hang on it: a recorded fact on AI pricing and market share, an open question over price versus share, and the judgment that any downshift in the software industry stays bounded. All three still await reassessment.
⑤ Does the standard change: No. The lesson here is the same as in the first entry: scheduled terms were read as a statement of the company's resolve. The first entry already changed the standard, so it isn't changed a second time here.
All of SecondSource's calls and their verdict dates → the judgment index
One more step: confirm by email
A confirmation email is on its way to the address just entered; clicking the link inside completes the subscription. Not in the inbox? Check spam or Promotions. Wrong address? Enter it again
Subscribe free — first issue lands tomorrow morning
Just an email address, unsubscribe anytime. Subscribing finishes with one click on the link in the confirmation email; if it is not in the inbox, check spam or Promotions.
More in this section
Checking the receipts: did Kimi K3's effective price fall to DeepSeek's level before October 15?
No new claim enters the ledger this week. This edition only goes back to one earlier call …
Does directors' liability make the White House AI accord binding? Gavin Baker's claim, checked against the accord's full text, Delaware case law and liability-insurance terms
The short version: Gavin Baker says directors' liability gives the accord binding force. O…
Is Anthropic's line 'We largely understand the risks today's models present' true? Miles Brundage's 'obviously false', checked against the same week's incident records
The short version: the receipts say unresolved. Read the line as a statement about the who…
Does Google's TPU beat Nvidia by up to 50% on performance per dollar? The claim, held up against the SemiAnalysis newsletter it came from
The short version: the receipts say unresolved. The 50% is not miscalculated. It holds at …